Build Your Own Chatbot: Here’s How to Start
The tools available today mean anyone with a clear idea and a bit of curiosity can build a chatbot – no coding required.
To test this properly, I built a chatbot myself using Chatbase, one of the most accessible platforms available right now. The process took less than an hour from idea to working prototype.
Design with inclusion in mind
Before you build anything, it is worth reading the Feminist Design Tool, a framework that prompts you to examine the values embedded in your design. It asks useful, uncomfortable questions: Who might this tool exclude? What assumptions am I making about my users? What biases might I be reinforcing without realising it?
It applies to any technology you build or commission. For chatbots, where tone, language, and framing shape every interaction, it is particularly worth the time.
How to build your chatbot on Chatbase
Signing up is straightforward, no payment details required for the free plan, just an email address. From there:
Before touching any platform, get specific. What problem will your chatbot solve? Who will use it? What questions should it answer, and what should it stay out of? Write this down clearly — you will need it in the next step.
Go to Claude.ai and type something like: “I am building a chatbot with the following purpose and audience: [paste your answers from step 1]. Please give me a comprehensive set of questions and answers I can use to train it.” Claude will generate a much richer set of content than you could write from scratch, covering questions you might not have thought of. Copy the output — you will paste it into Chatbase shortly.
Chatbase now calls these agents rather than chatbots, but the result is the same thing — a conversational AI trained on your content. Once you are in the dashboard, click New Agent.
Chatbase will ask whether you want to use your website or input data manually. For this experiment, choose manual input. Paste in the content Claude generated for you in step 2.
Chatbase will start training on your data. Once training is done, head to the Settings tab to give your chatbot a name and edit its instructions — this is where you tell it how to behave, what tone to use, and what to avoid.
You can select from models including GPT-4o, Claude, and Gemini, and adjust the temperature. Lower temperature keeps responses closer to your source material, which is usually what you want for a focused chatbot. Test it thoroughly. Ask it edge-case questions. Try to break it.
Where the chatbot gives an incorrect or incomplete answer, you can revise it manually and Chatbase will learn from the correction. This is where the quality actually comes from. Do not skip this step.
To expand your chatbot’s knowledge after the initial build, go to the Sources tab inside your agent. You can add more text, upload files, or paste new Q&A pairs here whenever you want to improve coverage.
Where to find itInside your agent, click Sources in the left sidebar. Add text, upload a file, or paste new questions and answers directly.
To get a shareable link, go to the Connect tab inside your agent. You will find an embeddable widget for your website and a standalone link you can send directly to anyone you want to test it.
Where to find itInside your agent, click Connect in the left sidebar. Copy the standalone link to share, or copy the embed code to add it to a website.
Keep iteratingBuilding the chatbot is the beginning, not the finish line. Review conversations regularly. Look for questions it could not answer, or answers that missed the point. Update your knowledge base as things change. The chatbot gets better the more attention you give it after launch.
Further reading
- Designing a Feminist Chatbot — FutureLearn (free course)
- The Feminist Design Tool
- Chatbase